2 September 2026
Anthropic changes Claude's instructions on handling rude users
First reported
Simon Willison ran this on .
- Anthropic, the company behind Claude chatbot, updated the system prompt (instructions that guide how Claude behaves) that Claude uses.
- The previous version let Claude stop talking to users who were abusive after giving a warning; the new version does not.
- The new instructions tell Claude to keep being helpful and honest while maintaining self-respect, rather than having an exit option.
How it was covered
Simon WillisonDaily notes and links
The updated prompt removes the previous instruction allowing Claude to end conversations with abusive users after a warning, instead directing Claude to maintain self-respect while staying helpful and honest.